lsfhuihuiff / MusicTI_AAAI2024

" Music Style Transfer with Time-Varying Inversion of Diffusion Models"
MIT License
35 stars 4 forks source link

Some questions of image size and the embdding. #1

Open spiralanch opened 8 months ago

spiralanch commented 8 months ago

Thanks for the amazing work, the result is awesome on https://lsfhuihuiff.github.io/MusicTI/. Still have some problems that not well understand below, looking forward to your reply, thanks!

  1. How i get the mel-spectrograms image? Can you offer the script, please? Is the same picture size between different length of audio, for example, 5 seconds audio and 60 seconds audio have the same image size?
  2. Whether to offer the pretrained embdding for faster testing the result? “--embedding_path /path/to/logs/trained_model/checkpoints/”
lsfhuihuiff commented 8 months ago

1.As mentioned in README, you can refer to Riffusion for the conversion between spectrograms and audio. Spectrograms of different durations correspond to spectrograms of different lengths, while the width remains consistent. 2.Each style audio corresponds to one model, making it difficult to upload the models. I will attempt to upload 1-2 models.

spiralanch commented 8 months ago

Thanks your reply. The model files on Google drive need permission, request access all ready. It would be better if permissions could be opened.

1.As mentioned in README, you can refer to Riffusion for the conversion between spectrograms and audio. Spectrograms of different durations correspond to spectrograms of different lengths, while the width remains consistent. 2.Each style audio corresponds to one model, making it difficult to upload the models. I will attempt to upload 1-2 models.

lsfhuihuiff commented 8 months ago

They have been made public.

spiralanch commented 8 months ago

They have been made public.

The model files are only 5 kB per item, errror happened when loading the *.pt, please recheck if the files is right.

lsfhuihuiff commented 8 months ago

I apologize for any errors that may have occurred during the upload. I have already corrected them.

farahhuifanyang commented 8 months ago

I apologize for any errors that may have occurred during the upload. I have already corrected them.

Hey, guy! See if u could fix this anyway. image